aevynt/jupimind-datasets

JupiMind数据集是一个用于JupiMind训练工作流程的标准化数据集快照集合。主要资产是用于`aevynt/JupiMind-Dense-J1R-Tokenizer-24k`的预训练混合数据集。数据集包含越南语和英语文本，主要用于文本生成任务。数据集格式为JSONL，每条记录包含一个文本字段。数据集包含约1,710,000条序列，估计token预算约为7B。数据来源包括FineWeb2-HQ、越南语精选数据集、维基百科越南语版等。数据集的许可证因来源不同而异，没有统一的许可证。

The JupiMind datasets repository stores normalized dataset snapshots used for JupiMind training workflows. The primary asset at the moment is the pretraining mix built for `aevynt/JupiMind-Dense-J1R-Tokenizer-24k`. The dataset includes Vietnamese and English texts, primarily for text-generation tasks. The format is JSONL with one record per line as `{"text": "..."}`. It contains approximately 1,710,000 sequences with an estimated token budget of about ~7B tokens. Source datasets include FineWeb2-HQ, Vietnamese curated dataset, Vietnamese Wikipedia, etc. The licensing varies by upstream source, with no single blanket content license applied to the mixed dataset.